Platinum: Path-Adaptable LUT-Based Accelerator Tailored for Low-Bit Weight Matrix Multiplication

TL;DR

Platinum is a novel ASIC accelerator that enhances low-bit weight matrix multiplication efficiency by adaptive path switching and reduced LUT overhead, significantly improving speed and energy efficiency for quantized neural networks.

Contribution

It introduces a path-adaptable LUT-based accelerator with offline LUT construction and adaptive execution paths for ternary and general bit-serial weights, reducing overhead and boosting performance.

Findings

Achieves up to 73.6x speedup over existing accelerators.

Reduces energy consumption by up to 32.4x.

Operates efficiently within a 0.96mm2 chip area.

Abstract

The rapid scaling of large language models demands more efficient hardware. Quantization offers a promising trade-off between efficiency and performance. With ultra-low-bit quantization, there are abundant opportunities for results reuse, and thus it can be boosted with lookup tables (LUTs) based acceleration. However, existing LUT-based methods suffer from computation and hardware overheads for LUT construction, and rely solely on bit-serial computation, which is suboptimal for ternary-weight networks. We propose Platinum, a lightweight ASIC accelerator for integer weight mixed-precision matrix multiplication (mpGEMM) using LUTs. Platinum reduces LUT construction overhead via offline-generated construction paths and supports both general bit-serial and optimized ternary-weight execution through adaptive path switching. On BitNet b1.58-3B, Platinum achieves up to 73.6x, 4.09x, and 2.15x speedups over SpikingEyeriss, Prosperity, and 16-thread T-MAC (CPU), respectively, along with energy reductions of 32.4x, 3.23x, and 20.9x, all within a 0.96mm2 chip area. This demonstrates the potential of LUT-based ASICs as efficient, scalable solutions for ultra-low-bit neural networks on edge platforms.